Sensors for coding agents: computational vs inferential, fast vs slow

## Spark

Fowler's three-part sensors trilogy (Three More Static Code Analysis Sensors, Maintainability Sensors for Coding Agents) clustered in the 2026-05-23 digest alongside Husain's Evals Skills and the user's own work: `woodland-generators#322` (Vale tuning for technical docs) and `alunduil-chezmoi#164` (fabrication, doc, and tool-discovery rules). Same taxonomy showing up independently across reading and repo work.

## Why it could be interesting

The post isn't "I configured Vale and ESLint." The claim: agent-assisted dev needs an explicit sensor taxonomy — computational vs inferential × fast (continuous) vs slow (drift-detecting) — and personal fleets converge on it independently of any single source. The artifact would be a 2×2 of the fleet's own sensors, mapped to where each lives (pre-commit, CI, scheduled review, ad-hoc skill), with the unfilled quadrants honestly named.

## Open questions

- [ ] Which sensors are already in place across alunduil/* and dungeon-studio, and where are the gaps?
- [ ] Inferential slow-cadence sensors (security review, data-handling review, dependency freshness) — what's a sustainable cadence at scale-of-one?
- [ ] Does the taxonomy hold across language ecosystems (Haskell, TypeScript, chezmoi shell, Terraform)?
- [ ] How does the CLAUDE.md fleet rollout (issue #172 and siblings) fit — rule surface or sensor?

## Source material

- https://read.readwise.io/read/01ks7c79kt24t1fmfj5t5t6vw3 — Three More Static Code Analysis Sensors (Fowler)
- https://read.readwise.io/read/01ks24wx4zkhr0g9s0q09csjn8 — Maintainability Sensors for Coding Agents (Fowler)
- https://read.readwise.io/read/01ks0dq17aajwz28dc12g5a54h — Evals Skills for Coding Agents (Husain)
- https://read.readwise.io/read/01ks0df0w5many2fga5d8t9a6y — Design-First Collaboration (Garg)
- https://github.com/alunduil/woodland-generators/pull/322 — Vale tuning for technical docs
- https://github.com/alunduil/alunduil-chezmoi/pull/164 — fabrication, doc, and tool-discovery rules